Preferred Tempo and Low-Audio-Frequency Bias

0 downloads 0 Views 4MB Size Report
May 29, 2018 - activity to determine the tempo or beat frequency of the music. ... Keywords: auditory, rhythm, tempo induction, musical beat, biomimetic model.
ORIGINAL RESEARCH published: 29 May 2018 doi: 10.3389/fnins.2018.00349

Preferred Tempo and Low-Audio-Frequency Bias Emerge From Simulated Sub-cortical Processing of Sounds With a Musical Beat Nathaniel J. Zuk 1*, Laurel H. Carney 1,2 and Edmund C. Lalor 1,2,3,4 1

Department of Biomedical Engineering, University of Rochester, Rochester, NY, United States, 2 Department of Neuroscience, University of Rochester Medical Center, Rochester, NY, United States, 3 Del Monte Institute for Neuroscience, University of Rochester Medical Center, Rochester, NY, United States, 4 Trinity Centre for Bioengineering, Trinity College Dublin, Dublin, Ireland

Edited by: Mounya Elhilali, Johns Hopkins University, United States Reviewed by: Michael Schwartze, Maastricht University, Netherlands Hugo Merchant, Universidad Nacional Autónoma de México, Mexico Aniruddh Patel, Tufts University, United States Jonathan Cannon contributed to the review of Aniruddh Patel *Correspondence: Nathaniel J. Zuk [email protected] Specialty section: This article was submitted to Neuromorphic Engineering, a section of the journal Frontiers in Neuroscience Received: 13 January 2018 Accepted: 07 May 2018 Published: 29 May 2018 Citation: Zuk NJ, Carney LH and Lalor EC (2018) Preferred Tempo and Low-Audio-Frequency Bias Emerge From Simulated Sub-cortical Processing of Sounds With a Musical Beat. Front. Neurosci. 12:349. doi: 10.3389/fnins.2018.00349

Prior research has shown that musical beats are salient at the level of the cortex in humans. Yet below the cortex there is considerable sub-cortical processing that could influence beat perception. Some biases, such as a tempo preference and an audio frequency bias for beat timing, could result from sub-cortical processing. Here, we used models of the auditory-nerve and midbrain-level amplitude modulation filtering to simulate sub-cortical neural activity to various beat-inducing stimuli, and we used the simulated activity to determine the tempo or beat frequency of the music. First, irrespective of the stimulus being presented, the preferred tempo was around 100 beats per minute, which is within the range of tempi where tempo discrimination and tapping accuracy are optimal. Second, sub-cortical processing predicted a stronger influence of lower audio frequencies on beat perception. However, the tempo identification algorithm that was optimized for simple stimuli often failed for recordings of music. For music, the most highly synchronized model activity occurred at a multiple of the beat frequency. Using bottom-up processes alone is insufficient to produce beat-locked activity. Instead, a learned and possibly top-down mechanism that scales the synchronization frequency to derive the beat frequency greatly improves the performance of tempo identification. Keywords: auditory, rhythm, tempo induction, musical beat, biomimetic model

INTRODUCTION When we spontaneously tap our feet to music, we are “feeling the beat.” A musical beat is frequently defined by the effect it has on motor entrainment (Patel, 2010; London, 2012), and it is often identified as the fundamental level in the metrical hierarchy for keeping time (Lerdahl and Jackendoff, 1983). Many cultures have music with a beat, and the presence of beat-based music is highly related to communal dance (Savage et al., 2015). Clearly, perceiving the beat is key to the perception of music. In many genres of music, musical beats often, but not always, occur at isochronous intervals (London, 2012). Previous models have simulated the perception of isochronous beats using an

Frontiers in Neuroscience | www.frontiersin.org

1

May 2018 | Volume 12 | Article 349

Zuk et al.

Sub-cortical Coding of Musical Beats

sounds, the exact frequency range of the bias, and the influence of subcortical processing on the bias, is still unclear. Separately, several groups have developed “tempo-induction” algorithms that identify the tempo of musical recordings (for review see Gouyon et al., 2006; McKinney et al., 2007). These algorithms typically consist of three stages: identify onsets in the music, determine the pattern and speed of those onsets, and determine the tempo based on several representations of these factors (ex. Elowsson and Friberg, 2015). While some of these algorithms use processes that are similar to the auditory system (ex. Scheirer, 1998), none have been built on biomimetic models of auditory processing that simulate the neural activity produced by stages of auditory processing below the cortex. This processing is important because beat perception is based on the neural activity projected to the cortex. Both physiological modulation tuning and the inherent randomness of neural signals present in realistic auditory processing could affect beat perception in real music. Here, we developed a model that determines the tempo of recordings of music based on the simulated neural activity of the auditory nerve and amplitude modulation tuning in the brainstem and midbrain. We hypothesized that physiologically plausible synaptic processing, which results in amplitude modulation tuning in the midbrain, can impose a preferred tempo near 100 BPM (London, 2012). We also hypothesized that innate processing in the auditory nerve can explain our lowaudio-frequency bias for timing musical beats. Lastly, we quantify the strength of neural synchronization to musical beats in musical recordings and assess different ways in which the beat frequency may be inferred based on sub-cortical processing.

internal clock (Povel and Essens, 1985), pattern matching (Rosenthal, 1992; Parncutt, 1994), an internal resonator (van Noorden and Moelants, 1999), or a bank of neural oscillators (Large et al., 2015). These models often compute the beat frequency of discrete pulses, although a few have used annotated performances as input (ex. Rosenthal, 1992) or “onset signals” computed from cochlear-like filtering of audio signals (Scheirer, 1998; Large, 2000). Using electroencephalography and magnetoencephalography, it has been shown that cortical activity time-locks to the perceived beat for simplistic stimuli (Snyder and Large, 2005; Iversen et al., 2009; Nozaradan et al., 2011, 2012; Fujioka et al., 2012, 2015; Tierney and Kraus, 2015; Tal et al., 2017; but see Henry et al., 2017). Yet multiple stages of processing occur prior to cortical processing, each of which could affect the placement of musical beats. Even for basic acoustic events, human subjects are biased to tapping to beats at inter-onset intervals of 500 to 700 ms (Parncutt, 1994), equivalent to a tempo range of 85 to 120 BPM. This range encompasses the “indifference interval” (Fraisse, 1963; London, 2012) for which subjects tap naturally at a regular rhythm (Semjen et al., 1998), discriminate tempi best (Drake and Botte, 1993), and can best replicate the duration of the interval (Stevens, 1886; Woodrow, 1934) (for review see Fraisse, 1963; Patel, 2010; London, 2012). This range also overlaps the range of tempi for a large proportion of dance music, which centers on 450 to 600 ms for intervals between beats, or equivalently 100 to 133 BPM (van Noorden and Moelants, 1999). However, an explanation for this optimal range of tempi is unclear. Motor entrainment plays a role in this bias since subjects tap naturally within this range, but it does not completely explain the optimization observed in studies that do not involve motor entrainment. Modulation tuning in the sub-cortical central nervous system would affect the synchronization strength of neural activity to isochronous acoustic events, which in turn could influence the preferred tempo. Additionally, there is some evidence that our perception of musical beats is biased to certain ranges of audio frequencies. Listeners’ ratings of “groove” in music, a subjective quality related to how much people want to move to the music, is correlated with the fluctuation in energy in low frequency (